The BLKSIZE parameter is associated with the DD statement.

The purpose of BLKSIZE is to improve processing efficiency by reducing the number of physical I/Os by grouping records. The penalty however is an increase in memory as the size of the data buffers is increased to match the BLKSIZE.

If the program does not contain BLKSIZE=0 or equivalent (BLOCK CONTAINS 0 RECORDs in COBOL) then BLKSIZE on the DD statement and the file label will be overridden.

If the program or DD statement specify a BLKSIZE less than the actual size of blocks in the data set, a System 001 Abend will occur. (IBM Utilities raise a Wrong Length Record condition, but it is the same thing in reality.)

BLKSIZE (Block Size)

This stands for Block Size and prior to MVS/ESA 4.3 the BLKSIZE parameter had to be coded as a sub-parameter of DCB, as in:

// DCB=(BLKSIZE=0,LRECL=nnnnn,RECFM=FB)

BLKSIZE has subsequently been propogated to full parameter status, and does not need to be part of the DCB sub-parameter list, so the following is valid:

// BLKSIZE=0,LRECL=nnnnn,RECFM=FB

Both methods are still valid so BLKSIZE can be coded either way and historical JCL is likely to include the former method.

Recommendations:

For input data sets it is best to omit the BLKSIZE from the DD statement, allowing the system to determine the length from the program or file label.
For output data sets it is best to code BLKSIZE=0 to allow System Determined BLKSIZE (SDB) to calculate an optimum size.

There is a commentary on optimum BLKSIZE, and click here to calculate an optimum BLKSIZE.